Classification and characterization of research output from the social sciences and humanities based on textual metadata

The classification of scientific research plays an important role in structuring the research landscape and in supporting the monitoring and evaluation of science. By assigning publications, journals, and organizations to disciplines, classification systems help make sense of an increasingly complex body of scholarly output. However, widely used international citation databases, such as Web of Science, are known to insufficiently cover the Social Sciences and Humanities (SSH). To address these limitations, national bibliographic databases have been developed, including the Flemish Academic Bibliographic Database for the Social Sciences and Humanities (VABB-SHW). While these databases provide broader SSH coverage, they also face challenges related to limited citation metadata, multilingual publications, diverse output types, and incomplete textual information. These characteristics make traditional ways of positioning publications into disciplines impractical. Against this background, this dissertation explores whether recent developments in Natural Language Processing (NLP) can be used to support text-based, publication-level classification of SSH research. The research combines conceptual reflection on academic disciplines and classification with empirical studies of text-based approaches. These studies show that the performance of text-based classification methods depends on disciplinary characteristics, publication types, and the availability of training data. In its application to VABB-SHW, the dissertation examines how disciplines are defined and operationalized in the database and what the implications and limitations of these definitions are. It further explores the possibilities opened by publication-level classification for extending insights derived from the database and for complementing existing classification practices. Although a large part of the research focuses on this Flemish SSH database, many of the questions addressed and insights developed are relevant beyond this specific case, both for the study of text-based classification of science and for the development of other local and national SSH databases.